A fabrication process of silicon chips is optimized for the type of transistors used for an application. Some of the key processes prevalent today are for logic, Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), flash memory and analog transistors. It is very desirable to have logic and memory, such as DRAM, flash memory, or SRAM on the same chip. Currently, one reason for not combining these example technologies on the same chip is due to the high degree of connectivity (or maximum number of wires) needed between the building blocks that are built with logic and memory on the same chip.
Higher connectivity between circuits yields high performance and low cost. But, it is not practical to integrate logic and memory on the same chip due to major differences in the fabrication processes. So, presently, manufacturers make one chip with logic and another with memory and connect them together with Ball Grid Arrays (BGA's). Connectivity of two chips using a BGA is significantly lower than of connectivity on a single silicon or other material chip. A BGA and its connectivity are described briefly herein to show a mismatch in the connectivity of BGA's and the chip.
FIG. 1A shows a chip 101 with a BGA 102, where each ball 104 of the BGA 102 is a semi-spherical ball 104 disposed on a surface 105 of the chip 101. An area 103 of the BGA 102 is described below in reference to FIG. 1B.
FIG. 1B is a sectional view of the chip 101 of FIG. 1A at AA′. The area 103 shows that a ball 104 of the BGA 102 is raised above the surface 105 of the chip 101.
FIG. 2A is an enlarged top view of the area 103, and FIG. 2B is a sectional view of the area 103, where a section BB′ of FIG. 2A is shown in FIG. 2B. Referring to both FIGS. 2A and 2B, a silicon substrate 201 has an insulator 202, such as SiO2. A first metal or aluminum connection 206 is formed on the insulator 202 for providing a chip interconnect. A wire 208 connected to an internal interconnect (not shown) of the chip 101 is connected to a pad 207. An insulator 209 covers the first metal connection 206. A contact hole 203 allows a second metal connection 204 to connect with the first metal connection 206. A semi hemisphere ball 205 is disposed on top of the second metal connection 204. An array of semi-hemisphere balls 205 is formed on the chip 101.
FIG. 3A shows two chips 301 and 302 flipped upside down and placed on a ceramic substrate 303 with a plated interconnect (not shown). FIG. 3B is a sectional view along section AA′ of FIG. 3A. Balls 304 on the chips 301, 302 are placed on metal or copper posts 305 on the substrate 303. By heating the whole assembly (i.e., the chips 301, 302 and ceramic substrate 305), the balls 304 are soldered or fused to the posts 305. As shown, the posts 305 have a pitch 306.
FIG. 3C is an enlarged view of an area 307 of FIG. 3B. A ceramic substrate 308 has a copper post 309 that connects to another post (not shown) through copper plated multilevel wiring, as known in the art. Silicon substrate 310 supports an insulation layer 311. A metal connection 312 is covered by an insulator layer 313, which has a contact hole 314. Another metal connection 315, may be formed of copper or gold, for example, and is placed on top of the contact hole 314. A metal semi-hemispheric ball 316 is placed or otherwise set on the metal connection 315. The chip 301 and substrate 308 are aligned so that the ball 316 is on top of the post 309. By heating the assembly of FIG. 3C, the ball 316 is soldered or fused to the post 309. Once soldered, the BGA assembly of two interconnected chips 301 and 302 and the substrate 303 is complete.
The pitch 306 of the posts 305 determines connectivity of the BGA. State of the art pitch ‘p’ is 4 mil or 100 microns. The connectivity ‘C’ is defined as a maximum number of wires that can connect two chips 301 and 302. C is equal to 1/p2, which is 1E04/cm2 for BGA's. On the other hand, connectivity of a state of the art silicon chip is significantly larger. In a silicon chip, if a feature size of a CMOS process is ‘F’, then the number of via connections on the top layer of the chip determines the connectivity of the chip. The connectivity C for a chip is 1/(2F2). For a state of the art CMOS process, the feature size F is 100 nm. So the maximum connectivity for a chip is 1E12/cm2.
The gap between the connectivity of a BGA and that of a chip is very large. In addition to connectivity, the capacitances at the output of the BGA, ceramic substrate posts and plated interconnects are very high and cause significant delay in signals going from one chip 301 to another 302. These delays limit the clock rate ‘f’ at which signals can travel between logic and memory chips. The best clock rate possible for BGA's is 100 MHz at commonly used TTL signal levels as compared to clock rates ‘f’ of 1 GHz within a 90 nm CMOS chip. Lower clock rates and low connectivity result in lower bandwidth for a signal to be transferred between logic on one chip 301 (FIG. 3B) and memory on another chip 302 (FIG. 3B). If logic and memory are on the same chip, data bandwidth can be increased by a ratio:R=(fchip×Cchip)/(fBGA×CBGA),
where
fchip is a clock rate possible in a within a chip (e.g., 4 GHz in 100 nm CMOS),
Cchip is connectivity within a chip (e.g., 1E12/cm2),
fBGA is clock rate possible within a BGA assembly (e.g., 100 MHz), and
CBGA is connectivity of a BGA assembly (e.g., 1E04/cm2). Hence, ratio R=1E08. Such a high ratio can open up many areas of applications, which are not possible today through the use of BGA technology or other chip interconnect technologies.